4 * Copyright IBM, Corp. 2007
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
14 #include <qemu-common.h>
16 #include "virtio-blk.h"
17 #include "block_int.h"
22 typedef struct VirtIOBlock
28 char serial_str[BLOCK_SERIAL_STRLEN + 1];
32 static VirtIOBlock *to_virtio_blk(VirtIODevice *vdev)
34 return (VirtIOBlock *)vdev;
37 /* store identify data in little endian format
39 static inline void put_le16(uint16_t *p, unsigned int v)
44 /* copy to *dst from *src, nul pad dst tail as needed to len bytes
46 static inline void padstr(char *dst, const char *src, int len)
49 *dst++ = *src ? *src++ : '\0';
52 /* setup simulated identify data as appropriate for virtio block device
54 * ref: AT Attachment 8 - ATA/ATAPI Command Set (ATA8-ACS)
56 static inline void virtio_identify_template(struct virtio_blk_config *bc)
58 uint16_t *p = &bc->identify[0];
59 uint64_t lba_sectors = bc->capacity;
61 memset(p, 0, sizeof(bc->identify));
62 put_le16(p + 0, 0x0); /* ATA device */
63 padstr((char *)(p + 23), QEMU_VERSION, 8); /* firmware revision */
64 padstr((char *)(p + 27), "QEMU VIRT_BLK", 40); /* model# */
65 put_le16(p + 47, 0x80ff); /* max xfer 255 sectors */
66 put_le16(p + 49, 0x0b00); /* support IORDY/LBA/DMA */
67 put_le16(p + 59, 0x1ff); /* cur xfer 255 sectors */
68 put_le16(p + 80, 0x1f0); /* support ATA8/7/6/5/4 */
69 put_le16(p + 81, 0x16);
70 put_le16(p + 82, 0x400);
71 put_le16(p + 83, 0x400);
72 put_le16(p + 100, lba_sectors);
73 put_le16(p + 101, lba_sectors >> 16);
74 put_le16(p + 102, lba_sectors >> 32);
75 put_le16(p + 103, lba_sectors >> 48);
78 typedef struct VirtIOBlockReq
81 VirtQueueElement elem;
82 struct virtio_blk_inhdr *in;
83 struct virtio_blk_outhdr *out;
84 struct virtio_scsi_inhdr *scsi;
86 struct VirtIOBlockReq *next;
89 static void virtio_blk_req_complete(VirtIOBlockReq *req, int status)
91 VirtIOBlock *s = req->dev;
93 req->in->status = status;
94 virtqueue_push(s->vq, &req->elem, req->qiov.size + sizeof(*req->in));
95 virtio_notify(&s->vdev, s->vq);
100 static int virtio_blk_handle_write_error(VirtIOBlockReq *req, int error)
102 BlockInterfaceErrorAction action = drive_get_onerror(req->dev->bs);
103 VirtIOBlock *s = req->dev;
105 if (action == BLOCK_ERR_IGNORE)
108 if ((error == ENOSPC && action == BLOCK_ERR_STOP_ENOSPC)
109 || action == BLOCK_ERR_STOP_ANY) {
114 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
120 static void virtio_blk_rw_complete(void *opaque, int ret)
122 VirtIOBlockReq *req = opaque;
124 if (ret && (req->out->type & VIRTIO_BLK_T_OUT)) {
125 if (virtio_blk_handle_write_error(req, -ret))
129 virtio_blk_req_complete(req, VIRTIO_BLK_S_OK);
132 static VirtIOBlockReq *virtio_blk_alloc_request(VirtIOBlock *s)
134 VirtIOBlockReq *req = qemu_mallocz(sizeof(*req));
139 static VirtIOBlockReq *virtio_blk_get_request(VirtIOBlock *s)
141 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
144 if (!virtqueue_pop(s->vq, &req->elem)) {
154 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
156 struct sg_io_hdr hdr;
162 * We require at least one output segment each for the virtio_blk_outhdr
163 * and the SCSI command block.
165 * We also at least require the virtio_blk_inhdr, the virtio_scsi_inhdr
166 * and the sense buffer pointer in the input segments.
168 if (req->elem.out_num < 2 || req->elem.in_num < 3) {
169 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
174 * No support for bidirection commands yet.
176 if (req->elem.out_num > 2 && req->elem.in_num > 3) {
177 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
182 * The scsi inhdr is placed in the second-to-last input segment, just
183 * before the regular inhdr.
185 req->scsi = (void *)req->elem.in_sg[req->elem.in_num - 2].iov_base;
186 size = sizeof(*req->in) + sizeof(*req->scsi);
188 memset(&hdr, 0, sizeof(struct sg_io_hdr));
189 hdr.interface_id = 'S';
190 hdr.cmd_len = req->elem.out_sg[1].iov_len;
191 hdr.cmdp = req->elem.out_sg[1].iov_base;
194 if (req->elem.out_num > 2) {
196 * If there are more than the minimally required 2 output segments
197 * there is write payload starting from the third iovec.
199 hdr.dxfer_direction = SG_DXFER_TO_DEV;
200 hdr.iovec_count = req->elem.out_num - 2;
202 for (i = 0; i < hdr.iovec_count; i++)
203 hdr.dxfer_len += req->elem.out_sg[i + 2].iov_len;
205 hdr.dxferp = req->elem.out_sg + 2;
207 } else if (req->elem.in_num > 3) {
209 * If we have more than 3 input segments the guest wants to actually
212 hdr.dxfer_direction = SG_DXFER_FROM_DEV;
213 hdr.iovec_count = req->elem.in_num - 3;
214 for (i = 0; i < hdr.iovec_count; i++)
215 hdr.dxfer_len += req->elem.in_sg[i].iov_len;
217 hdr.dxferp = req->elem.in_sg;
218 size += hdr.dxfer_len;
221 * Some SCSI commands don't actually transfer any data.
223 hdr.dxfer_direction = SG_DXFER_NONE;
226 hdr.sbp = req->elem.in_sg[req->elem.in_num - 3].iov_base;
227 hdr.mx_sb_len = req->elem.in_sg[req->elem.in_num - 3].iov_len;
228 size += hdr.mx_sb_len;
230 ret = bdrv_ioctl(req->dev->bs, SG_IO, &hdr);
232 status = VIRTIO_BLK_S_UNSUPP;
234 hdr.resid = hdr.dxfer_len;
235 } else if (hdr.status) {
236 status = VIRTIO_BLK_S_IOERR;
238 status = VIRTIO_BLK_S_OK;
241 req->scsi->errors = hdr.status;
242 req->scsi->residual = hdr.resid;
243 req->scsi->sense_len = hdr.sb_len_wr;
244 req->scsi->data_len = hdr.dxfer_len;
246 virtio_blk_req_complete(req, status);
249 static void virtio_blk_handle_scsi(VirtIOBlockReq *req)
251 virtio_blk_req_complete(req, VIRTIO_BLK_S_UNSUPP);
253 #endif /* __linux__ */
255 static void virtio_blk_handle_write(VirtIOBlockReq *req)
257 BlockDriverAIOCB *acb;
259 acb = bdrv_aio_writev(req->dev->bs, req->out->sector, &req->qiov,
260 req->qiov.size / 512, virtio_blk_rw_complete, req);
262 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
266 static void virtio_blk_handle_read(VirtIOBlockReq *req)
268 BlockDriverAIOCB *acb;
270 acb = bdrv_aio_readv(req->dev->bs, req->out->sector, &req->qiov,
271 req->qiov.size / 512, virtio_blk_rw_complete, req);
273 virtio_blk_req_complete(req, VIRTIO_BLK_S_IOERR);
277 static void virtio_blk_handle_output(VirtIODevice *vdev, VirtQueue *vq)
279 VirtIOBlock *s = to_virtio_blk(vdev);
282 while ((req = virtio_blk_get_request(s))) {
283 if (req->elem.out_num < 1 || req->elem.in_num < 1) {
284 fprintf(stderr, "virtio-blk missing headers\n");
288 if (req->elem.out_sg[0].iov_len < sizeof(*req->out) ||
289 req->elem.in_sg[req->elem.in_num - 1].iov_len < sizeof(*req->in)) {
290 fprintf(stderr, "virtio-blk header not in correct element\n");
294 req->out = (void *)req->elem.out_sg[0].iov_base;
295 req->in = (void *)req->elem.in_sg[req->elem.in_num - 1].iov_base;
297 if (req->out->type & VIRTIO_BLK_T_SCSI_CMD) {
298 virtio_blk_handle_scsi(req);
299 } else if (req->out->type & VIRTIO_BLK_T_OUT) {
300 qemu_iovec_init_external(&req->qiov, &req->elem.out_sg[1],
301 req->elem.out_num - 1);
302 virtio_blk_handle_write(req);
304 qemu_iovec_init_external(&req->qiov, &req->elem.in_sg[0],
305 req->elem.in_num - 1);
306 virtio_blk_handle_read(req);
310 * FIXME: Want to check for completions before returning to guest mode,
311 * so cached reads and writes are reported as quickly as possible. But
312 * that should be done in the generic block layer.
316 static void virtio_blk_dma_restart_bh(void *opaque)
318 VirtIOBlock *s = opaque;
319 VirtIOBlockReq *req = s->rq;
321 qemu_bh_delete(s->bh);
327 virtio_blk_handle_write(req);
332 static void virtio_blk_dma_restart_cb(void *opaque, int running, int reason)
334 VirtIOBlock *s = opaque;
340 s->bh = qemu_bh_new(virtio_blk_dma_restart_bh, s);
341 qemu_bh_schedule(s->bh);
345 static void virtio_blk_reset(VirtIODevice *vdev)
348 * This should cancel pending requests, but can't do nicely until there
349 * are per-device request lists.
354 /* coalesce internal state, copy to pci i/o region 0
356 static void virtio_blk_update_config(VirtIODevice *vdev, uint8_t *config)
358 VirtIOBlock *s = to_virtio_blk(vdev);
359 struct virtio_blk_config blkcfg;
361 int cylinders, heads, secs;
363 bdrv_get_geometry(s->bs, &capacity);
364 bdrv_get_geometry_hint(s->bs, &cylinders, &heads, &secs);
365 memset(&blkcfg, 0, sizeof(blkcfg));
366 stq_raw(&blkcfg.capacity, capacity);
367 stl_raw(&blkcfg.seg_max, 128 - 2);
368 stw_raw(&blkcfg.cylinders, cylinders);
369 blkcfg.heads = heads;
370 blkcfg.sectors = secs;
372 virtio_identify_template(&blkcfg);
373 memcpy(&blkcfg.identify[VIRTIO_BLK_ID_SN], s->serial_str,
374 VIRTIO_BLK_ID_SN_BYTES);
375 memcpy(config, &blkcfg, sizeof(blkcfg));
378 static uint32_t virtio_blk_get_features(VirtIODevice *vdev)
380 VirtIOBlock *s = to_virtio_blk(vdev);
381 uint32_t features = 0;
383 features |= (1 << VIRTIO_BLK_F_SEG_MAX);
384 features |= (1 << VIRTIO_BLK_F_GEOMETRY);
386 features |= (1 << VIRTIO_BLK_F_SCSI);
388 if (strcmp(s->serial_str, "0"))
389 features |= 1 << VIRTIO_BLK_F_IDENTIFY;
394 static void virtio_blk_save(QEMUFile *f, void *opaque)
396 VirtIOBlock *s = opaque;
397 VirtIOBlockReq *req = s->rq;
399 virtio_save(&s->vdev, f);
402 qemu_put_sbyte(f, 1);
403 qemu_put_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
406 qemu_put_sbyte(f, 0);
409 static int virtio_blk_load(QEMUFile *f, void *opaque, int version_id)
411 VirtIOBlock *s = opaque;
416 virtio_load(&s->vdev, f);
417 while (qemu_get_sbyte(f)) {
418 VirtIOBlockReq *req = virtio_blk_alloc_request(s);
419 qemu_get_buffer(f, (unsigned char*)&req->elem, sizeof(req->elem));
427 VirtIODevice *virtio_blk_init(DeviceState *dev, DriveInfo *dinfo)
430 int cylinders, heads, secs;
431 static int virtio_blk_id;
434 s = (VirtIOBlock *)virtio_common_init("virtio-blk", VIRTIO_ID_BLOCK,
435 sizeof(struct virtio_blk_config),
436 sizeof(VirtIOBlock));
438 s->vdev.get_config = virtio_blk_update_config;
439 s->vdev.get_features = virtio_blk_get_features;
440 s->vdev.reset = virtio_blk_reset;
443 if (strlen(ps = (char *)drive_get_serial(s->bs)))
444 strncpy(s->serial_str, ps, sizeof(s->serial_str));
446 snprintf(s->serial_str, sizeof(s->serial_str), "0");
447 s->bs->private = dev;
448 bdrv_guess_geometry(s->bs, &cylinders, &heads, &secs);
449 bdrv_set_geometry_hint(s->bs, cylinders, heads, secs);
451 s->vq = virtio_add_queue(&s->vdev, 128, virtio_blk_handle_output);
453 qemu_add_vm_change_state_handler(virtio_blk_dma_restart_cb, s);
454 register_savevm("virtio-blk", virtio_blk_id++, 2,
455 virtio_blk_save, virtio_blk_load, s);